RUIN THEORY for ENERGY-EFFICIENT RESOURCE ALLOCATION in UAV-ASSISTED CELLULAR NETWORKS 3945 Cellular Networks and Meet 5G NR Performance Requirements

IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 6, JUNE 2021 3943 Ruin Theory for Energy-Efﬁcient Resource Allocation in UAV-Assisted Cellular Networks Aunas Manzoor , Kitae Kim , Shashi Raj Pandey , S. M. Ahsan Kazmi , Nguyen H. Tran , Senior Member, IEEE, Walid Saad , Fellow, IEEE, and Choong Seon Hong , Senior Member, IEEE

Abstract— Unmanned aerial vehicles (UAVs) can provide an I. INTRODUCTION effective solution for improving the coverage, capacity, and HE use of unmanned aerial vehicles (UAVs) can enable a the overall performance of terrestrial wireless cellular net- wide range of smart city applications, ranging from drone works. In particular, UAV-assisted cellular networks can meet T the stringent performance requirements of the fifth generation delivery to surveillance and monitoring [1]–[3]. Recently, new radio (5G NR) applications. In this article, the problem the use of UAVs has greatly increased in wireless-networking of energy-efficient resource allocation in UAV-assisted cellular applications to provide coverage and capacity enhancement to networks is studied under the reliability and latency constraints the ground wireless networks [4]. The flexibility, autonomy, of 5G NR applications. The framework of ruin theory is and ease of deployment of UAVs render them suitable to be employed to allow solar-powered UAVs to capture the dynamics of harvested and consumed energies. First, the surplus power a part of the future wireless networks. In wireless networking of every UAV is modeled, and then it is used to compute the applications, UAVs can have various roles that range from fly- probability of ruin of the UAVs. The probability of ruin denotes ing base stations (BSs) ([3]) to backhaul nodes ([5]) and users the vulnerability of draining out the power of a UAV. Next, of the cellular network ([3], [6], [7]). Therefore, by leveraging the probability of ruin is used for efficient user association line-of-sight (LoS) communication at high altitudes as well with each UAV. Then, power allocation for 5G NR applications is performed to maximize the achievable network rate as the dynamic placement of UAVs at desired locations and using the water-filling approach. Simulation results demonstrate within a required time [8], the use of UAVs as flying BSs that the proposed ruin-based scheme can enhance the flight can play a significant role in boosting the capacity of cellular duration up to 61% and the number of served users in a networks [9]. For example, in [10], the authors proposed UAV flight by up to 58%, compared to a baseline SINR-based an UAV-assisted heterogeneous cellular network (HetNet) to scheme. meet the communication demands in emergencies for public Index Terms— 5G new radio (5G NR), user association, safety. The authors in [11] used optimal transport theory to energy efficiency, power allocation, ruin theory, surplus process, enable UAVs to provide communication services to ground unmanned aerial vehicles, URLLC. users while optimizing their flight time. Moreover, the authors of [12] proposed an efficient UAV BSs in coexistence with a Manuscript received September 15, 2020; revised January 18, 2021 and terrestrial network. In particular, when wireless connectivity is February 23, 2021; accepted February 27, 2021. Date of publication March 9, 2021; date of current version June 16, 2021. This work was supported by needed in difficult and costly deployment locations, UAVs can the National Research Foundation of Korea(NRF) grant funded by the Korea provide low-cost and low-power alternatives and complement government(MSIT) (No. 2020R1A4A1018607) and by the Institute of Infor- conventional small cell BSs (SBSs). For instance, a joint posi- mation & communications Technology Planning & Evaluation (IITP) grant funded by the Korea government(MSIT) (No. 2019-0-01287, Evolvable Deep tioning and user association problem was introduced in [13], Learning Model Generation Platform for Edge Computing). The associate where UAVs were used as a replacement for terrestrial SBSs. editor coordinating the review of this article and approving it for publication The authors in [14] used circle packing theory to efficiently was H. Suraweera. (Corresponding author: Choong Seon Hong.) Aunas Manzoor, Kitae Kim, Shashi Raj Pandey, and Choong Seon Hong deploy multiple UAVs that provide maximum coverage for are with the Department of Computer Science and Engineering, Kyung Hee users. The authors in [15] proposed the deployment of UAVs University, Yongin 446-701, South Korea (e-mail: [email protected]; to power the sensors in an IoT network. The authors in [16] [email protected]; [email protected]; [email protected]). S. M. Ahsan Kazmi is with the Department of Computer Science and proposed a 3D point process-based UAV deployment where Creative Technologies, University of the West of England (UWE), Bristol the UAVs with multiple antennas provide downlink coverage BS16 1QY, U.K., and also with the Department of Computer Science and to the ground users. The authors in [17] proposed a 3D Engineering, Kyung Hee University, Yongin 446-701, South Korea (e-mail: [email protected]). deployment scheme for the UAVs while guaranteeing the Nguyen H. Tran is with the School of Computer Science, The University of safety distance between UAVs in mmWave network. More- Sydney, Sydney, NSW 2006, Australia (e-mail: [email protected]). over, UAVs are a very promising solution to the problem of Walid Saad is with Wireless@VT, Bradley Department of Electrical and Computer Engineering, Virginia Tech, Blacksburg, VA 24061 USA, and also connectivity in occasionally crowded areas, such as stadiums with the Department of Computer Science and Engineering, Kyung Hee or open-air shows. Owing to these benefits of UAVs in University, Yongin 446-701, South Korea (e-mail: [email protected]). wireless communications, it is envisioned that future wireless Color versions of one or more figures in this article are available at https://doi.org/10.1109/TCOMM.2021.3064968. cellular networks [4] will be UAV-assisted because they can Digital Object Identifier 10.1109/TCOMM.2021.3064968 complement their terrestrial infrastructure with flying UAV 0090-6778 © 2021 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: Kyunghee Univ. Downloaded on June 17,2021 at 05:23:30 UTC from IEEE Xplore. Restrictions apply. 3944 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 6, JUNE 2021

BSs. Meanwhile, to design efficient UAV-assisted cellular energy-efficient solutions for UAV communication networks networks, it is necessary to address various challenges that have been proposed to address this problem [31]–[36]. For range from network modeling to optimization and resource instance, the authors in [31] developed a non-cooperative management [3]. game to optimize the beaconing periods among competing The use of UAVs is particularly meaningful for deliver- drones. In this way, the energy consumption in individual ing 5G new radio (NR) applications [18]. Enhanced mobile drones is optimized in a distributed manner. The authors broadband (eMBB) users require high data rates for 3D in [32] proposed a spectrum and energy-efficient scheme movies, gaming and AR/VR services. UAVs can promise such for UAV-enabled relay network, in which the UAV path is high data rates by exploiting the LoS communication [19]. optimized by allocating the communication time slots between Since UAVs have already been used for data collection from source and destination nodes. In [33], the authors proposed Internet of Things (IoT) networks [20], they can perform an energy-aware power allocation scheme in the UAV-assisted well for massive machine type communication (mMTC). edge networks while utilizing the internet of vehicles for Moreover, to meet the reliability and latency demands of the computation offloading. The authors in [34] developed a ultra-reliable and low-latency communication (URLLC) appli- UAV placement scheme to maximize the served users while cations, the LoS communication, and optimal positioning consuming the minimum transmit power. The authors in [35] features of UAVs can be promising. Recently, the use of UAVs designed a UAV communication scheme to optimize the UAV has been proposed in delay-sensitive and mission critical appli- trajectory while minimizing the energy consumption during cations [21]–[23]. The authors in [21] proposed a dynamic tra- the UAV flight. The authors in [36] proposed an energy effi- jectory control algorithm to optimize the delay and throughput ciency scheme in downlink (DL) and uplink (UL) decoupled in the UAV-aided networks. In [22], the authors developed a access network. The authors in [37] proposed a downlink URLLC channel model to ensure the delay-sensitive delivery energy transfer from the UAVs to the ground IoT devices, of critical control information from the ground station to the and the uplink information transfer from IoT devices to the UAV. The authors in [23] used a URLLC-enabled UAV to UAVs. develop a relay system for the delay-sensitive and ultra-reliable In order to improve energy efficiency in UAV-assisted communication. The authors in [24] used UAV as a relaying wireless networks, it is possible to harness the possibility node between the controller and robot when there is no LoS of energy harvesting (particularly using renewable sources link available between them. The location and power of the such as solar energy), a technical challenge that has not UAV is optimized to meet the URLLC reliability requirements. been considered in the aforementioned works [30]–[35]. By The authors in [25] studied the use of ultra-reliable low latency employing energy harvesting, the same UAVs can be used communications for controlling UAVs and allowing them to to serve the cellular users for a comparatively longer dura- avoid collisions. The authors in [26] discussed the URLLC tion to increase the air time of UAV. In this regard, there communication for IoT networks while analyzing the security exist very few studies [38]–[41] that discussed UAVs with challenges in IoT. However, most of these prior works focused energy-harvesting capabilities for wireless networking. In [38], on the use of UAVs to incorporate mMTC or URLLC services, the authors studied the problem of energy management in rather than consolidating all of the 5G services. Therefore, heterogeneous networks that integrate solar-powered drones. we propose the UAV-assisted cellular networks to serve the In particular, the approach in [39] is used to optimize the cellular users while incorporating eMBB, URLLC and mMTC trips of the drones at specific locations so as to minimize services of 5G NR. the total energy consumption in the network. The authors User association and power allocation schemes for the in [40] addressed the problem of energy limitations in drones UAV-based networks have been proposed in the literature, by developing a wireless power recharging system which is e.g., in [27]–[29] and [30]. For instance, the authors in [27] on renewable energy harvesting. In [41], a comprehensive proposed sum power minimization scheme to optimize user analysis of the energy harvesting was performed for UAVs association and power allocation in mobile edge computing from the solar and wind sources. The problem of signal- network (MEC). The authors used an iterative algorithm to to-noise-ratio outage minimization for the ground users was solve the formulated problem by decomposing it into three formulated to optimize the transmit power and flight duration subproblems. The authors in [28] proposed a joint user asso- of UAVs. However, since all the aforementioned works assume ciation and power control scheme for UAV-enabled cellular energy harvesting in UAV networks that are deployed to networks. The authors applied matching-based scheme to form assist traditional wireless networks, the study of UAV-enabled coalition game among the UAVs. The authors in [29] proposed networks to satisfy the requirements of heterogeneous user a machine learning based resource allocation scheme for traffic remains largely overlooked. In practice, it is imperative mutli-UAV network. The authors in [30] proposed to jointly to consider the coexistence between UAV networks with an solve the user association and power allocation problem in underlaid next-generation cellular infrastructure incorporating UAV-enabled MEC networks using successive convex approx- eMBB, URLLC and mMTC services. imation scheme. B. Contributions A. Energy Efficiency in UAV-Assisted Cellular Networks The main contribution of this article is a novel The efficient utilization of the limited onboard energy of a energy-efficient user association and power allocation UAV to serve cellular users is a significant challenge. Many framework for optimizing energy-constrained UAV-assisted

Authorized licensed use limited to: Kyunghee Univ. Downloaded on June 17,2021 at 05:23:30 UTC from IEEE Xplore. Restrictions apply. MANZOOR et al.: RUIN THEORY FOR ENERGY-EFFICIENT RESOURCE ALLOCATION IN UAV-ASSISTED CELLULAR NETWORKS 3945 cellular networks and meet 5G NR performance requirements. To accomplish this goal, we first formulate a joint user association and power allocation problem to achieve the maximum rate for enhanced mobile broadband (eMBB) users in the network under the constraints of URLLC quality-of-service (QoS) requirements and limited power of BSs. Using the framework of ruin theory [42], [43], we solve the user association problem by modeling the surplus power in every UAV BS (UBS) [44]. Here, surplus power is defined as the total residual power for a given UAV at any time instant. In particular, we determine the so-called probability of ruin of the surplus power in each UBS and then use it for user association with UBSs. The adoption of ruin theory to model the surplus power in the UAVs helps to efficiently capture two opposing power flows that include: (a) the periodic power harvested in the form of regular premiums, and (b) the power allocated to the associated Fig. 1. System model of UAV-assisted cellular network. cellular users in the form of claims. In this regard, conventional stochastic optimization cannot properly capture the dynamics number of users during a UAV flight compared to a of these opposing powers in UAVs. For instance, [35] pre- benchmark SINR-based scheme. sented an energy efficient trajectory design scheme for the UAVs. However, they do not consider the energy harvesting To the best of our knowledge, this is the first study that in their energy model. The authors in [45] studied the battery applies ruin theory to model the surplus power in UBSs for dynamics of the UAVs to capture the charging and discharging optimizing user association in UAV-assisted cellular networks. flows of battery power. In contrast to these schemes, ruin The rest of this article is organized as follows. Section II theory captures the charging and discharging flows of powers presents the system model and problem formulation. as well as the randomness in the transmission power of the In Section III, we develop a ruin theoretic framework to model UAVs. After obtaining the energy-efficient association of the the surplus UAV power which is used for the user association. cellular users with a particular BS based on the probability of In Section III-D, the proposed water-filling based power allo- ruin, the power allocation to the associated users is performed cation algorithm is developed; and numerical evaluation of the under 5G NR constraints. proposed framework are discussed in Section IV. Conclusions In summary, our key contributions include the following: are drawn in Section V. • To maximize the data rate of eMBB users while satisfying the latency and reliability constraints of URLLC users, II. SYSTEM MODEL AND PROBLEM FORMULATION we formulate a joint energy-efficient, user association, We consider a UAV-assisted cellular network that comprises and power allocation problem. The formulated problem of a single macro BS (MBS), a set S of S SBSs, and a set is difficult to solve because the problem is mixed integer U of U UBSs for the downlink communication. These BSs non-linear programming (MINLP). Therefore, we pro- can be compositely denoted by J = {0}∪S∪U,where pose a novel ruin-based user association and water-filling {0} denotes the index of the MBS. Every UAV has a limited algorithm for the power allocation to the associated power storage capacity to store ρ0 power at the start of a eMBB users. flight. Moreover, there is a solar energy-harvesting module • First, we model the surplus power in a UBS as function mounted on every UAV. The UAVs are deployed above the of three components: (a) the initial power in the UAVs ground cellular network at the corresponding locations denoted at the time of dispatch, (b) the regular harvested power by ru =(xu,yu,hu),foreveryUAVu ∈U[22]. We consider was obtained in the form of premiums from the solar the following three types of users in 5G NR environment: (a) a panels mounted on the UAVs, and (c) the transmission set Ke of Ke eMBB users, (b) a set Ku of Ku URLLC users, power from UAVs to the cellular users. From the surplus and (c) a set Km of Km mMTC users. The data obtained process, we determine the probability of ruin which is from all the IoT nodes of mMTC network is multiplexed in a further used for the user association. single frame. The data demand from this single frame is equal • After user association, we develop a new power allocation to one regular eMBB user demand. Therefore, without loss algorithm to allocate downlink power to the associated of generality, we remark that there is only one mMTC user cellular users. Power allocation to the URLLC users is denoted by index {0} in the network. All these cellular users initially performed under a reliability constraint and, then, are compositely denoted by K = {0}∪Ke ∪Ku,where{0} the water-filling scheme is used to allocate power to denotes the index of the mMTC user. eMBB users. We study the network for a short duration of time, and the • Simulation results demonstrate that UAV-assisted cellu- network topology is considered static during this period. For lar networks perform well to satisfy the stringent 5G the sake of exposition, we consider that the UAVs u ∈U are NR requirements. Moreover, we show that the proposed deployed at key locations ru in which the cellular demand ruin-based user association serves up to 58% more is highest. This assumption is aligned with the main use-case

Authorized licensed use limited to: Kyunghee Univ. Downloaded on June 17,2021 at 05:23:30 UTC from IEEE Xplore. Restrictions apply. 3946 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 6, JUNE 2021 of UAVs as BSs that service hotspot areas for a temporary among the various cellular users associated with BS j.1 Wj ωjk = È duration. We assume that determining the optimal locations A bandwidth xjk is allocated to the communication follows known algorithms such as in [14]. k∈K link between the BS j ∈J and cellular user k ∈K,where k∈K xjk denotes the total number of associated users with A. Transmission Model the BS j. xjk denotes the association variable given as: Due to the different propagation environments, the channel 1, if user k ∈Kis associated with BS j ∈J, gain of the UBSs differs from the ground SBSs and the MBS. xjk −δ /10 = In particular, the channel gain is given by: hjk =10 jk 0, otherwise. and is a function of the corresponding pathloss δjk.The The achievable rate of a cellular user k associated with the path-loss δuk between UAVs and ground users is composed of LoS NLoS BS j is given by the following: LoS path-loss, δuk and non-LoS path-loss, δuk , respectively given as follows: R x ω γ , jk = jk jk log (1 + jk) (8) πd f δLoS 4 uk L , uk =20log c + LoS (1) B. 5G NR Traffic Classification Model πd f δNLoS 4 uk L , As previously explained, the set of cellular users K in uk =20log c + NLoS (2) the network can be classified into two main categories based where f is the carrier frequency, and duk denotes the distance on the 5G NR traffic classification as K = Ku ∪Ke while K ∩K ∅ K between UAV u and cellular user k. Moreover, LLoS and u e = ,where u denotes the set of URLLC users K LNLoS are the average added losses for LoS and NLOS and e denotes the set of eMBB users. This means that any communication, respectively. The probability of existing a LoS user k can be classified into one of these two network user link between UAVs and ground users is given as follows [46]: groups Ku or Ke. The data obtained from all the IoT nodes of mMTC network is multiplexed in a single subframe 1 ms. LoS 1 Pruk = , (3) Thus, we consider the communication for the mMTC traffic 1+a exp[b( 180 tan−1 hu ) − a] π duk which is multiplexed into a single subframe and is dedicated ω where hu is the height of the UAV u,anda and b are environ- fixed bandwidth of j0 as shown in Fig. 2. Additionally, ment constants, respectively. The probability of non-LoS link we consider the downlink communication for the URLLC between UAVs and ground users is given as follows: and eMBB traffic. It can be observed from Fig. 2 that the arrival of URLLC traffic reduces the achievable rate of the NLoS − LoS. Pruk =1 Pruk (4) eMBB users because eMBB traffic could be stopped during Hence, the average path-loss between UAVs and ground users URLLC transmission. The achievable eMBB rate after serving xjk URLLC requests is given as follows: is given as follows: k ∈Ku LoS LoS NLoS NLoS δuk = Pruk δuk + Pruk δuk . (5) Djk (xjk,Pjk)= T-t xjk Rjk, ∀k ∈Ke, (9) Path-loss δjk, for the terrestrial BSs j ∈{{0}∪S} is given k ∈Ku by the following [47], [48]: where T and t denote the eMBB and URLLC TTIs respec- D δjk =15.3+37.6log(djk), (6) tively as shown in Fig. 2. jk represents the amount of data which can be communicated from BS j to the eMBB user d j where jk denotes the distance between BS and cellular user k T xjk during time while simultaneously serving k ∈Ku k. URLLC users. The signal-to-interference-plus-noise ratio (SINR), which is denoted by γij from the BS j to the cellular user k is given by the following: C. Problem Formulation P h We can now formulate an optimization problem to maximize jk jk γjk = 2 , (7) the network rate for eMBB users subject to the URLLC Pjkhjk + σjkωjk j∈J \{j} reliability and latency constraints, the power level constraints of each BS j ∈J, and the unique association of the cellular P where jk denotes the downlink transmission power from BS users with each BS, as follows: j ∈J to the user k ∈K, hjk denotes the channel gain, 2 and σjk denotes the noise power spectrum density at user max Djk (xjk,Pjk) , (10) x,P k [26]. Note that Pj k denotes the interference power from the j∈J k∈K neighboring BS j to the user k and is equal to Pjk ,where Pjk ≤ ρj, ∀j ∈J, (10a) k is another user that uses the same frequency band as user k∈K k. ωjk denotes the bandwidth allocated to the communication Pr (γjk ≥ ζ) ≥ (1 − ), ∀j ∈J, ∀k ∈Ku, (10b) link between the BS j ∈J and cellular user k ∈K. Each j ∈J BS is assigned a portion of the licensed bandwidth 1Note that the proposed resource allocation scheme can be efficiently used denoted by Wj. This bandwidth Wj is further divided equally for the multiple antennas on separate instances for each channel resource.

of propulsion power and the power allocation to the cellular users. The propulsion power ρh, consumed in the hovering state of the UAV is constant due to the fixed air speed, UAV height and other environmental variables. The power allocation to the set of cellular users associated with UAV u is denoted by k∈K Puk. The surplus power of UAV u at time instant τ,isgivenby: u ρu(τ)=ρ0 + ρ τ − Puk − ρh. (12) k∈K Note that the surplus power of every UAV has an upper bound ρmax which denotes the limited power storage capacity of the UAVs. In the absence of the solar energy source at night time or cloudy weather, the UAVs can be switched to other renewable energy sources (e.g., laser, UV, or fuel cells) [45], [49]. We now observe that this surplus power has bidirectional flows of power. The positive power flow is in the form of periodic harvested power and the negative power flow is in the form of power consumption on the Fig. 2. The frame structure of 5G NR traffic for the coexistence of eMBB, downlink communication from UAVs to cellular users and URLLC, and mMTC. mMTC is allocated a dedicated bandwidth. eMBB traffic is pre-schedules for eMBB TTI duration, T in each frame. URLLC traffic is propulsion power. Hence, an efficient system design should scheduled instantaneously for URLLC TTI duration, t. well-capture these power flows to improve the overall energy utilization. In this regard, conventional stochastic optimization requires full network knowledge to provide a centralized xjk = λu, ∀j ∈J, (10c) solution; hence, less efficient. Therefore, we use a stochastic k∈Ku process-based framework of ruin theory to address this chal- xjk =1, ∀k ∈K, (10d) lenge and model the surplus UAV power and the probability of j∈J ruin of UAV to formulate the stochastic optimization problem.

0 ≤ Pjk ≤ pmax, ∀j ∈J,k∈K, (10e) To that end, the probability of ruin at UAV represents the vulnerability of draining out a UAV’s power. In particular, it is xjk ∈{0, 1}, ∀j ∈J,k∈K. (10f) used to efficiently utilize the available energy of UAVs while The objective is to maximize the eMBB users’ sum-rate in performing user association with each UAV. the network. Based on the maximum power level ρj of each j BS , the constraint in (10a) limits the total power allocation III. USER ASSOCIATION AND POWER ALLOCATION: to all the associated users. (10b) ensures ultra-reliability by ARUIN-BASED SOLUTION maintaining a sufficient SINR level for the URLLC users A. Ruin Theory: Preliminaries above the threshold ζ with 1− confidence level. (10c) ensures low latency for URLLC users by strictly scheduling the arrived In the area of actuarial science [44], ruin theory is applied λu URLLC requests in the same slot. (10d) ensures the unique to express an insurer’s vulnerability of bankruptcy. This is association of eMBB user k with a single BS j. (10e) and performed by modeling the so-called surplus process which τ (10f) are the bounds for the decision variables, where pmax represents the insurer’s capital at a time instant and com- denotes the maximum power level that can be allocated for a prises two opposing cash flows: (a) the periodic income gained user. Note that, the upper bound pmax is helpful in reducing from regular insurance premiums, and (b) random claims. the interference from neighboring BSs. The following is the The insurer’s vulnerability of risk is determined from the power level, ρj for each BS j ∈J: probability of ruin which essentially represents the probability ⎧ of getting a negative surplus. As explained earlier, we will ⎨⎪ρj(τ), for j ∈U, apply ruin theory to formulate the surplus power in every ρj = P0, for j =0, (11) UAV u at a time instant τ, and is further used to compute the ⎩⎪ P1, for j ∈S. probability of ruin. The number of cellular users associated with the UAV is then determined based on the probability of ruin of each UAV. D. UAV Energy Model Ruin theory is utilized to model the UAV surplus power Every UAV u ∈U has a certain power level ρu(τ) at time process which represents the random UAV power levels over instant τ, which comprises the initial power ρ0 stored in the time τ. The surplus UAV power depends on three essential UAV, t h e p ow e r ρu that is harvested by UAV u from renewable factors: a) launch power, b) premium power, and c) claim, energy sources within a unit time. Note that the unit time τ is that we define next. measured in seconds and is longer than the eMBB TTI T and Definition 1: The launch power, which is denoted by ρ0 is URLLC TTI t. The UAVs have power consumption in the form defined as the power stored in the UAV at the time of launch.

Authorized licensed use limited to: Kyunghee Univ. Downloaded on June 17,2021 at 05:23:30 UTC from IEEE Xplore. Restrictions apply. 3948 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 6, JUNE 2021

Definition 2: The premium power, which is denoted by ρu Appendix A), as follows: is defined as the constant power harvested by UAV u from the max ς Djk (xjk,Pjk) − ξ ψu(ρ0,τ), (16) renewable energy resource in a unit time τ. x,P j∈J k∈K u∈U Definition 3: The claims are defined as the random transmit power allocations to the cellular users associated with a UAV. Pjk ≤ ρj, ∀j ∈{{0}∪S}, (16a) The UAV surplus power, is composed of constant renewable k∈K γ ≥ζ ≥ − , ∀j ∈J, ∀k ∈K , power harvested at regular intervals, and the power dissipation Pr( jk ) (1 ) u (16b) on providing the communicatio n services to the associated xjk = λu, ∀j ∈J, (16c) cellular users, i.e., Puk. For simplicity, we ignore the k∈Ku k∈Ku constant hovering power ρh in the rest of the article. xjk =1, ∀k ∈K, (16d) Let, Su = Puk denote the compound random k∈Ku j∈J variable which is composed of two random variables Ku,and 0 ≤ Pjk ≤ p , ∀j ∈J,k∈K, (16e) Puk,whereKu denotes the number of users associated with max x ∈{ , }, ∀j ∈J,k∈K. the UAV u and Puk denotes the power allocated by the UAV jk 0 1 (16f) u k S to the associated cellular user . This compound variable u The objective is to maximize the sum-rate of the users μ is exponentially distributed with parameter . Consequently, in the network and minimize the probability of ruin for the the surplus power given in (12) can be formally defined as surplus power in the UAVs, where ς and ξ are normalizing follows: constants. Before proceeding to the solution, we first simplify u the problem by embedding (16c) in the objective function. ρu(τ)=ρ0 + ρ τ − Su. (13) This is done as follows: The surplus in (13), which represents the UAV power at (16c) is an equality constraint which ensures the immediate time instant, τ, is utilized to find the finite-time probability of scheduling of λu number of arrived URLLC users in the same ruin defined as follows [42]. time slot t. This scheduling is performed by associating the Definition 4: The finite-time probability of ruin is defined URLLC user k with the BS j delivering best SINR γjk .The as the probability of getting a negative surplus at any time resultant eMBB rate after URLLC scheduling is expressed as instant s during a finite time τ. The finite-time probability of follows: ruin is mathematically given by: Djk (xjk,Pjk)=(T-tλu) Rjk, ∀k ∈Ke, (17)

ψ(ρ0,τ)=Pr[ρu(s) < 0,forsomes as 0

Authorized licensed use limited to: Kyunghee Univ. Downloaded on June 17,2021 at 05:23:30 UTC from IEEE Xplore. Restrictions apply. MANZOOR et al.: RUIN THEORY FOR ENERGY-EFFICIENT RESOURCE ALLOCATION IN UAV-ASSISTED CELLULAR NETWORKS 3949 xjk =1, ∀k ∈K, (19b) Algorithm 1 User Association Algorithm j∈J 1: Input: J, K, Pjk, ρj ; ≤ Pjk ≤ p , ∀j ∈J,k∈K, ∗ 0 max (19c) 2: Initialize: xjk =0; xjk ∈{0, 1}, ∀j ∈J,k∈K. (19d) 3: Step 1: 4: Compute ψu(ρ0,τ) from (15); It can be observed that problem (III-B) is an MINLP, which 5: Compute ηjk from (21); is difficult to solve using exhaustive search and branch and 6: for k =1to K do bound techniques, particularly for a large network. Further- 7: Select single BS j with max ηjk; more, the binary variable xjk makes the problem combi- j∈J national which may require exponential-complexity to solve 8: end for using exhaustive search for the number of users. To avoid the 9: Step 2: j J difficulty, we convert the problem into simple sub-problems 10: for =1to do; P ρ which are solved for each variable separately. In particular, 11: Initialize = j; P ≥ to get a near optimal solution, we first fix the power allocation 12: while 0 do 13: Find max γjk; and solve the problem of user association by employing k∈K ∗ ruin-based heuristic approach. Then, we solve the power allo- 14: Update xjk =1,andP = P − Pjk; 15: Remove max γjk from SINR vector γjk; cation problem using the optimal ruin-based user association k∈K solution. 16: end while 17: end for C. Ruin-Based User Association With UAVs The first sub-problem solves xjk to estimate the possible we present a ruin-based heuristic approach in Algorithm 1 number of users which can be associated with each UAV. This that results to a near optimal solution. The inputs are the estimation is performed by modeling the surplus power of a following: the total number of BSs, J, the total number of UAV using the probability of ruin. After fixing Pjk, the user network users, K, fixed power allocation Pjk, allocated to each association sub-problem is given by: user k, and the total power bound, pj of BS j. In Algorithm 1, max ς D (xjk,Pjk) − ξ ψu(ρ0,τ), (20) first, the probability of ruin, ψu(ρ0,τ) is computed in step x jk j∈J k∈K u∈U 4 by using the initial power storage ρ0, the harvested power ρu and the random claim arrivals. The probability of ruin is xjk =1, ∀k ∈K, (20a) η j∈J further used to compute jk from (22) in step 5. Then, every cellular user, k, selects the BS j with the maximum value xjk ∈{0, 1}, ∀j ∈J,k ∈K. (20b) of ηjk in step 7. Next, the users are sorted according to the For the sake of energy-efficiency, the user association xjk, SINR values and are associated with the corresponding BS with every UAV u is performed based on its probability of under the constraints of power level ρj of each BS in step ∗ ruin, ψu(ρ0,τ), obtained from (15). This is done by limiting 13. The association xjk and the remaining BS power P is the number of associated users with each UAV based on the updated in step 14. The algorithm terminates when either P probability of ruin. Meanwhile, when the probability of ruin is zero or all the cellular users are associated. This algorithm of a UAV is high, fewer users are associated to that UAV gives a near-optimal solution with the complexity of O(N) and vice versa. One meaningful approach for user association for performing user association using the probability of ruin. is to associate the cellular user k, with that BS which is We will show the optimality gap in the simulation results in providing better SINR level γjk. This SINR-based approach Section IV, where it will be observed that the UAV with less is not energy-efficient and, hence, it is not suitable for the surplus power will have a high probability of ruin. Hence, problem of power constrained user association with the UAVs. fewer users will be associated with that UAV. We propose an approach that combines SINR and probability of ruin-based approach in the following definition. D. Power Allocation to eMBB Users Definition 5: To incorporate both the SINR and the proba- From the previous section, we obtain the optimal associated bility of ruin in the UAV user association problem, we intro- ∗ users denoted by xjk. After performing the ruin-based optimal duce a factor, ηjk, defined as follows: user association, the next step is to solve the power allocation ηjk := α(1 − ψu(ρ0,τ))γjk, (21) problem. Thus, the updated value of the achievable rate for the set of the associated eMBB users is given by: where the term (1 − ψu(ρ0,τ)) denotes the probability of ∗ survival and α is a control factor. Rjk =(T-tλu) xjkωjk log (1 + γjk) . (22) It can be observed that the factor ηjk consolidates both the The proof of equivalence of (19) and (23) is given in SINR and the probability of ruin. Therefore, the cellular user Appendix C. The sub-problem for the power allocation to is associated with that UBS that provides a better SINR to the eMBB users is given as follows: cellular user and has a relatively less probability of ruin. The user association problem (20) is combinatorial and max Rjk, (23) P difficult to solve for the large network. To solve the problem, j∈J k∈Ke

Authorized licensed use limited to: Kyunghee Univ. Downloaded on June 17,2021 at 05:23:30 UTC from IEEE Xplore. Restrictions apply. 3950 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 6, JUNE 2021

Algorithm 2 Iterative Ruin-Based Resource Allocation Algo- rithm 1: Input: J, K, Pjk, ρj, ηjk; 2: Set uniform power to each BS and user pair {j,k} i.e. Pjk[0] = ρj/K; 3: Compute γjk at each user k from the BS j; 4: Initialize: t =0; 5: Associate and allocate power to URLLC users using (18) ; t ≤ T ∗ ≥ 6: while max or 0 do ∗ 7: xjk[t] = argmax Djk; x j∈J k∈K ∗ 8: Allocate power Pjk to eMBB users using (24); 9: Update ∗ 10: Pjk[t]=Pjk; ∗ 11: = Pjk[t] − Pjk[t − 1]; 12: t = t +1; 13: end while

∗ s.t. Pjk ≤ ρj − Pjk , ∀j ∈{{0}∪S}, (23a) k∈Ke k ∈Ku

0 ≤ pjk ≤ pmax, ∀j ∈J,k∈Ke. (23b) Problem (23) is convex because the objective function and Fig. 3. Systematic diagram of the iterative algorithm. all of the constraints are convex. To solve this problem, we use a variant of the water-filling algorithm where the power, Pjk maximum power, while the leftover power is allocated to the allocated to a user is limited by a threshold, pmax,andthe other users by accordingly adjusting the water level, λj .The leftover power is allocated to the other low-gain network third user is allocated power using (24) based on the available users as shown in Fig. 4. The following optimal solution is channel gain, θjk, while the fourth user is not allocated any obtained: power because of a very small channel gain. Proposition 2: The optimal power allocation P ∗ is jk To solve (20), we present an iterative association and power expressed as: allocation approach in Algorithm 2. First, a uniform power is x∗ ω + ∗ jk jk 1 allocated to all the user-BS pairs. Then, the users maximizing Pjk =min pmax, − , ∀j ∈J,k ∈Ke, λj θjk the network rate under maximum power bounds are selected until the algorithm iteration time t is less than the maxi- (24) mum iteration threshold Tmax, or the algorithm convergence ∗ where θjk is the channel gain for the user k from BS j defined parameter is greater than the convergence threshold 0. as: Then, the power allocation for the selected users is performed h using the water-filling algorithm. Fig. 3 shows the systematic θ jk , ∀j ∈J,k ∈K . jk = 2 e 1+ Pjkhjk + ωjkσ diagram and flowchart of the iterative algorithm for the asso- j∈J \{j} ciation and power allocation. Note that this algorithm adopts (25) water-filling algorithm with the complexity of O(N.M).The joint algorithm complexity of the proposed algorithm compris- We note that the maximum power that can be allocated ing of both ruin-based algorithm and water-filling algorithm p λ∗ to a user is limited by max. From (24), we get j which is O(N.M). The algorithm gives a near-optimal solution and denotes the optimal water level chosen such that the following we show the optimality gap in the next section. condition is satisfied. ∗ ∗ Pjk = ρj − Pjk , ∀j ∈J. (26) IV. SIMULATION RESULTS k∈Ke k ∈Ku For our simulation, we consider a geographical area Proof: To get (26), we get the KKT conditions by com- of 4000 m × 4000 m square. MBS is deployed in the center at puting the Lagrangian function and other necessary condtions a fixed location, whereas 10 SBSs and 5 UAVs are uniformly of (25). By concurrently solving the KKT conditions we deployed in the area. The cellular users are randomly located obtain (26). in the geographical area. Statistical results are averaged over The maximum power bound pmax helps to serve more users several runs of random locations of cellular users, SBSs and by allocating the leftover power to low channel gain users. This UAVs. Other simulation parameters are given in table I. is performed by adjusting the water level, λj, using (26). This Fig. 5 shows the network topology which comprises a single is illustrated in Fig. 4 where the first two users are allocated MBS and uniformly deployed SBSs, UBSs and the cellular

Fig. 6. Network rate vs. number of cellular users in the network. Fig. 4. An illustration of the water-ﬁlling algorithm.

TABLE I SIMULATION PARAMETERS.

Fig. 7. Total surplus power of all UAVs at different time instants.

a better per-user rate when compared with the terrestrial network. For example, when the number of users in the network is about 75, the UAV-assisted network achieves about 40% more rate when compared with the terrestrial network. This is due to the LoS communication link between UAV and ground users which deliver better SINR as compare to the non-LoS link of the terrestrial network. From Fig. 6, we can also observe that the network rate reduces as the number of users in the network increases. This is due to the limited power resources at BSs which are insufficient to serve all the users in the network. We also observe that the decline in network rate is very slow. This is due to the optimal user association where the users with better SINR are associated with the BS. Fig. 7 illustrates how the surplus power per UAV varies Fig. 5. Network topology consisting of MBS, SBSs, UBSs and the cellular users. over time during a single flight of UAV. We show the power allocation to the associated users for 100 TTIs. The optimization problem is solved by running the ruin-based users. The UBSs are deployed at a height of 200 m above the association algorithm and water-filling power allocation ground level. algorithm for each TTI. The initial UAV surplus power at the Fig. 6 demonstrates the per-user network rate as a function time of launch t =0is set to be 100 Watts. By controlling the of the number of users. The network rate is calculated after number of the associated users through the probability of ruin, solving the optimization problem for 5 UAVs and 5 SBSs. up to 52% higher level of the UAV surplus power is achieved These results indicate that the UAV-assisted network achieves when compared with the non-ruin approach. By preserving the

Authorized licensed use limited to: Kyunghee Univ. Downloaded on June 17,2021 at 05:23:30 UTC from IEEE Xplore. Restrictions apply. 3952 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 6, JUNE 2021

Fig. 8. Number of total users associated with all UAVs vs. the probability Fig. 11. Plot of convergence. of ruin.

Fig. 12. Network rate vs. number of cellular users in the network.

Fig. 9. Comparison of ruin and SINR-based approach for UAV flight time and number of served users. Fig. 8 demonstrates how the number of UAV-associated cellular users varies with the probability of ruin. We compare the proposed ruin-based scheme with a baseline SINR-based scheme, which associates the cellular users with the UAVs to give the best SINR without consider- ing the surplus power of the UAVs. By decreasing the number of associated cellular users with the UAVs, it can be observed that the proposed ruin-based scheme tries to preserve UAV power with an increase in the probability of ruin. The non-associated users are offloaded to other UAVs with less probability of ruin or to the terrestrial network. Fig. 9 shows the flight time in units of TTIs and the number of served users during a single flight of UAV for the SINR-based and ruin-based approaches. Fig. 9 demonstrates that the ruin-based approach enhances the flight time of UAV by offloading the users to terrestrial network. As a result, Fig. 10. Network rate vs. number of cellular users in the network. exploiting the energy harvested during the enhanced flight duration, more number of cellular users are served by the UAV eventually. For instance, the flight duration of UAV surplus UAV power as shown by the significant differences in the ruin-based approach is 64 TTIs as compared to the in power drops at certain time instants, the ruin-based SINR-based approach. During 64 TTIs, additional energy is approach can serve more number of cellular users harvested which is further used to serve 93 more users during eventually. the UAV flight. Consequently, the proposed ruin-based scheme

Fig. 13. Scenarios: (a) sufficient transmission power at the UAVs, (b) the high channel gain users are allocated maximum power level pmax and the low channel gain users are allocated power proportionally, and (c) all the users have very low channel gain as compared to the available power. can enhance the flight duration up to 61% and the number of case Fig. 13(c), the available power is insufficient; therefore, served users in a UAV flight by up to 58%, compared to a the power is allocated to each user based on the noise level baseline SINR-based scheme. while satisfying the water level feasibility. Fig. 10 shows the plot of a network for the two traffic classes of 5G NR, i.e., eMBB and URLLC against the number V. C ONCLUSION of network users. From this figure, we can see that the rate of URLLC traffic is not affected by increasing the number In this article, we have studied the UAV-assisted cellular of network users. Meanwhile, the eMBB rate significantly networks to enhance the cellular network capacity. We have decreases as the number of users in the network increases. formulated a joint optimization problem for the user asso- This is because the URLLC users are given priority over the ciation and power allocation for the 5G NR traffic classifi- eMBB users by allocating the resources to satisfy their latency cations. First, we have performed the user association and and reliability requirements. power allocation to the URLLC users under reliability and Fig. 11 shows the convergence of power allocation algo- latency constraints. Then, by utilizing the probability of ruin to rithm against the number of iterations. The power alloca- estimate the possible number of cellular users to be associated tion is performed to the associated set of users by the BS with each UAV, we have solved the user association problem. in an iterative manner. It can be observed that the algo- Based on the probability of ruin of the UAV, the cellular rithm converges after 30 number of iterations. With the traffic was offloaded by associating the cellular users with help of water-filling scheme, the iterative power allocation is the other UAVs or the terrestrial network. Then, we have performed. iteratively solved the power allocation problem using the Fig. 12 demonstrates the optimality gap between the pro- water-filling scheme. Simulation results have demonstrated posed and optimal solution. We use the Gurobi optimiza- the effectiveness of the proposed ruin-based energy-efficiency tion tool to obtain the optimal solution of the original scheme in terms of increasing the flight time of UAVs and the problem II-C. We show the optimality gap results for the number of served cellular users. small network containing up to 30 cellular users. It can be observed that there is no gap for a small network when there are up to 7 cellular users in the network. A very small and APPENDIX A negligible gap is observed as the number of cellular users PROOF OF EQUIVALENCE OF (10) AND (16) increases in the network. This gap is due to the upper bound (10a) is separated for the UBSs j ∈U as follows: T of maximum number of iterations max. This upper bound restricts the iterative algorithm to the near-optimal solution Pjk ≤ ρj, ∀j ∈U, (27) for the large network. k∈K Fig. 13 illustrates three possible cases of the water-filling ρ ρ ρt j algorithm for power allocation. We show the results for a where j = 0 + denoted the power level of UAV . different amount of powers to be allocated to the associated By rearranging (27), we get that cellular users. In the first case Fig. 13(a), there is sufficient ρ0 + ρt − Pjk ≥ 0, ∀j ∈U. (28) available power which is allocated to the associated users k∈K based on the proportional noise level. It means that the user with high noise level is allocated less power and vice versa. Inequality (28) refers to the positive surplus power for every In the second case Fig. 13(b), there is enough power available UAV which is the definition of the probability of ruin. There- that pmax is allocated to the low-noise users and high-noise fore, inequality (28) is equivalent to minimizing the probability users are allocated power based on the water level. In the third of ruin.

Authorized licensed use limited to: Kyunghee Univ. Downloaded on June 17,2021 at 05:23:30 UTC from IEEE Xplore. Restrictions apply. 3954 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 6, JUNE 2021

APPENDIX B [8] J. Chen and D. Gesbert, “Optimal positioning of flying relays for PROOF OF POWER ALLOCATION TO URLLC USERS wireless networks: A LOS map approach,” in Proc. IEEE Int. Conf. Commun. (ICC), Chengdu, China, May 2017, pp. 1–6. From (16b), Pr (γjk ≥ ζ) can be expressed as CDF [9] V. Sharma, M. Bennis, and R. Kumar, “UAV-assisted heterogeneous networks for capacity enhancement,” IEEE Commun. Lett., vol. 20, no. 6, Fγ (ζ).Soweget: jk pp. 1207–1210, Jun. 2016. [10] A. Merwaday and I. Guvenc, “UAV assisted heterogeneous networks Fγ (ζ) ≥ (1 − ), (29) jk for public safety communications,” in Proc. IEEE Wireless Commun. ζ Netw. Conf. Workshops (WCNCW), New Orleans, LA, USA, Mar. 2015, From here, we can easily get the following expression for pp. 329–334. ζ ≥F−1 (1 − ) [11] M. Mozaffari, W. Saad, M. Bennis, and M. Debbah, “Wireless com- γjk (30) munication using unmanned aerial vehicles (UAVs): Optimal transport theory for hover time optimization,” IEEE Trans. Wireless Commun., The maximization of the eMBB data rate under the con- vol. 16, no. 12, pp. 8052–8066, Dec. 2017. straint (16b) gives boundary solution γjk = ζ for the optimal [12] M. G. Khoshkholgh, K. Navaie, H. Yanikomeroglu, V. C. M. Leung, power allocated to the URLLC users. The optimal power is and K. G. Shin, “Coverage performance of aerial-terrestrial HetNets,” in Proc. IEEE 89th Veh. Technol. Conf. (VTC), Kuala Lumpur, Malaysia, given as follows: Apr. 2019, pp. 1–5. F −1 (1 − )(1 + I) [13] J. Plachy, Z. Becvar, P. Mach, R. Marik, and M. Vondra, “Joint ∗ γjk positioning of flying base stations and association of users: Evolutionary- Pjk = , (31) hjk based approach,” IEEE Access, vol. 7, pp. 11454–11463, Jan. 2019. [14] M. Mozaffari, W. Saad, M. Bennis, and M. Debbah, “Efficient 2 where I = Pjkhjk + ωjkσ . deployment of multiple unmanned aerial vehicles for optimal wire- j∈J \{j} less coverage,” IEEE Commun. Lett., vol. 20, no. 8, pp. 1647–1650, Aug. 2016. [15] O. Cetinkaya and G. V. Merrett, “Efficient deployment of UAV-powered APPENDIX C sensors for optimal coverage and connectivity,” in Proc. IEEE Wireless PROOF OF EQUIVALENCE OF (19) AND (23) Commun. Netw. Conf. (WCNC), Seoul, South Korea, May 2020, pp. 1–6. [16] C.-H. Liu, D.-C. Liang, M. A. Syed, and R.-H. Gau, “A 3D tractable The objective function of problem (19) is given as follows: model for UAV-enabled cellular networks with multiple antennas,” IEEE Trans. Wireless Commun., early access, Jan. 21, 2021, doi: ς Djk − ξ ψu(ρ0,τ). (32) 10.1109/TWC.2021.3051415. j∈J k∈K u∈U [17] Y. Zhu, G. Zheng, and M. Fitch, “Secrecy rate analysis of UAV-enabled mmWave networks using Matérn hardcore point processes,” IEEE J. Sel. Areas Commun., vol. 36, no. 7, pp. 1397–1409, Jul. 2018. It can be seen that the factor ξ u∈U ψu(ρ0,τ) is constant [18] M. Bennis, M. Debbah, and H. V. Poor, “Ultrareliable and low-latency when observed from the perspective of power allocation and wireless communication: Tail, risk, and scale,” Proc. IEEE, vol. 106, therefore can be ignored in the optimization problem (23). no. 10, pp. 1834–1853, Oct. 2018. Then after the URLLC association xjk according to Proposi- [19] W.-E. Chen, X.-Y. Fan, and L.-X. Chen, “A CNN-based packet classification of eMBB, mMTC and URLLC applications for 5G,” in Proc. Int. tion 1 and the ruin-based eMBB association xjk,theeMBB Conf. Intell. Comput. Emerg. Appl. (ICEA), Paris, France, Aug. 2019, data rate is given as follows: pp. 140–145. ∗ [20] M. Mozaffari, W. Saad, M. Bennis, and M. Debbah, “Mobile Rjk =(T-tλu) xjkωjk log (1 + γjk) , (33) unmanned aerial vehicles (UAVs) for energy-efficient Internet of Things communications,” IEEE Trans. Wireless Commun., vol. 16, no. 11, which is the objective function of the problem (23). pp. 7574–7589, Nov. 2017. [21] Z. M. Fadlullah, D. Takaishi, H. Nishiyama, N. Kato, and R. Miura, “A dynamic trajectory control algorithm for improving the communica- REFERENCES tion throughput and delay in UAV-aided networks,” IEEE Netw., vol. 30, no. 1, pp. 100–105, Jan. 2016. [1] R. W. Beard, T. W. McLain, D. B. Nelson, D. Kingston, and [22] H. Ren, C. Pan, K. Wang, Y. Deng, M. Elkashlan, and A. Nallanathan, D. Johanson, “Decentralized cooperative aerial surveillance using fixed- “Achievable data rate for URLLC-enabled UAV systems with 3-D chan- wing miniature UAVs,” Proc. IEEE, vol. 94, no. 7, pp. 1306–1324, nel model,” IEEE Wireless Commun. Lett., vol. 8, no. 6, pp. 1587–1590, Jul. 2006. Dec. 2019. [2] M. Mozaffari, A. T. Z. Kasgari, W. Saad, M. Bennis, and M. Debbah, [23] C. Pan, H. Ren, Y. Deng, M. Elkashlan, and A. Nallanathan, “Beyond 5G with UAVs: Foundations of a 3D wireless cellular net- “Joint blocklength and location optimization for URLLC-enabled UAV work,” IEEE Trans. Wireless Commun., vol. 18, no. 1, pp. 357–372, IEEE Commun. Lett. Jan. 2019. relay systems,” , vol. 23, no. 3, pp. 498–501, Mar. 2019. [3] M. Mozaffari, W. Saad, M. Bennis, Y.-H. Nam, and M. Debbah, “A tutorial on UAVs for wireless networks: Applications, challenges, [24] H. Ren, C. Pan, K. Wang, W. Xu, M. Elkashlan, and A. Nallanathan, and open problems,” IEEE Commun. Surveys Tuts., vol. 21, no. 3, “Joint transmit power and placement optimization for URLLC-enabled pp. 2334–2360, 3rd Quart., 2019. UAV relay systems,” IEEE Trans. Veh. Technol., vol. 69, no. 7, [4] W. Saad, M. Bennis, and M. Chen, “A vision of 6G wireless systems: pp. 8003–8007, Jul. 2020. Applications, trends, technologies, and open research problems,” IEEE [25] K. Wang, C. Pan, H. Ren, W. Xu, L. Zhang, and A. Nallanathan, “Packet Netw., vol. 34, no. 3, pp. 134–142, May 2020. error probability and effective throughput for ultra-reliable and low- [5]E.Kalantari,M.Z.Shakir,H.Yanikomeroglu, and A. Yongacoglu, latency UAV communications,” IEEE Trans. Commun., vol. 69, no. 1, “Backhaul-aware robust 3D drone placement in 5G+ wireless net- pp. 73–84, Jan. 2021. works,” in Proc. IEEE Int. Conf. Commun. Workshops (ICC Workshops), [26] H. Ren, C. Pan, Y. Deng, M. Elkashlan, and A. Nallanathan, “Resource Paris, France, May 2017, pp. 109–114. allocation for secure URLLC in mission-critical IoT scenarios,” IEEE [6] U. Challita, W. Saad, and C. Bettstetter, “Interference management Trans. Commun., vol. 68, no. 9, pp. 5793–5807, Sep. 2020. for cellular-connected UAVs: A deep reinforcement learning approach,” [27] L. Yang, J. Chen, M. O. Hasna, and H.-C. Yang, “Outage performance IEEE Trans. Wireless Commun., vol. 18, no. 4, pp. 2125–2140, of UAV-assisted relaying systems with RF energy harvesting,” IEEE Apr. 2019. Commun. Lett., vol. 22, no. 12, pp. 2471–2474, Dec. 2018. [7] R. Amer, W. Saad, and N. Marchetti, “Mobility in the sky: Performance [28] M. Sami and J. N. Daigle, “User association and power control for and mobility analysis for cellular-connected UAVs,” IEEE Trans. Com- UAV-enabled cellular networks,” IEEE Wireless Commun. Lett.,vol.9, mun., vol. 68, no. 5, pp. 3229–3246, May 2020. no. 3, pp. 267–270, Mar. 2020.

[29] Z. Chang, W. Guo, X. Guo, and T. Ristaniemi, “Machine learning-based Aunas Manzoor received the M.S. degree in resource allocation for multi-UAV communications system,” in Proc. electrical engineering, with a specialization in IEEE Int. Conf. Commun. Workshops (ICC Workshops), Dublin, Ireland, telecommunication, from the National University of Jun. 2020, pp. 1–6. Science and Technology, Pakistan, in 2015. He is [30] Z. Yang, C. Pan, K. Wang, and M. Shikh-Bahaei, “Energy effi- currently pursuing the Ph.D. degree with the Depart- cient resource allocation in UAV-enabled mobile edge computing net- ment of Computer Science and Engineering, Kyung works,” IEEE Trans. Wireless Commun., vol. 18, no. 9, pp. 4576–4589, Hee University, South Korea. His research interest Sep. 2019. includes applying analytical techniques for resource [31] S. Koulali, E. Sabir, T. Taleb, and M. Azizi, “A green strategic activity management in mobile cellular networks. scheduling for UAV networks: A sub-modular game perspective,” IEEE Commun. Mag., vol. 54, no. 5, pp. 58–64, May 2016. [32] J. Zhang, Y. Zeng, and R. Zhang, “Spectrum and energy efficiency maximization in UAV-enabled mobile relaying,” in Proc. IEEE Int. Conf. Commun. (ICC), Chengdu, China, May 2017, pp. 1–6. Kitae Kim received the B.S. and M.S. degrees in [33] L. Zhang, Z. Zhao, Q. Wu, H. Zhao, H. Xu, and X. Wu, “Energy-aware computer science and engineering from Kyung Hee dynamic resource allocation in UAV assisted mobile edge computing University, Seoul, South Korea, in 2017 and 2019, over social Internet of vehicles,” IEEE Access, vol. 6, pp. 56700–56715, respectively, where he is currently pursuing the Oct. 2018. Ph.D. degree in computer science and engineering. [34] M. Alzenad, A. El-Keyi, F. Lagum, and H. Yanikomeroglu, “3-D place- His research interests include SDN/NFV, wireless ment of an unmanned aerial vehicle base station (UAV-BS) for energy- networks, unmanned aerial vehicle communications, efficient maximal coverage,” IEEE Wireless Commun. Lett., vol. 6, no. 4, and machine learning. pp. 434–437, Aug. 2017. [35] Y. Zeng and R. Zhang, “Energy-efficient UAV communication with trajectory optimization,” IEEE Trans. Wireless Commun., vol. 16, no. 6, pp. 3747–3760, Jun. 2017. [36] Y. Shi, E. Alsusa, and M. W. Baidas, “Energy-efficient decoupled access scheme for cellular-enabled UAV communication systems,” IEEE Syst. Shashi Raj Pandey received the B.E. degree in J., early access, Jan. 22, 2021, doi: 10.1109/JSYST.2020.3046689. electrical and electronics, with a specialization in [37] Y. Zhu, G. Zheng, K.-K. Wong, and T. Dagiuklas, “Spectrum and energy communication, from Kathmandu University, Nepal, efficiency in dynamic UAV-powered millimeter wave networks,” IEEE in 2013. He is currently pursuing the Ph.D. degree Commun. Lett., vol. 24, no. 10, pp. 2290–2294, Oct. 2020. in computer science and engineering with Kyung [38] D. Vincent, P. S. Huynh, L. Patnaik, and S. S. Williamson, “Prospects Hee University, South Korea. He served as a Net- of capacitive wireless power transfer (C-WPT) for unmanned aerial work Engineer with Huawei Technologies Nepal vehicles,” in Proc. IEEE PELS Workshop Emerg. Technol., Wireless Co., Pvt., Ltd., Nepal, from 2013 to 2016. His Power Transf. (Wow), Montréal, QC, Canada, Jun. 2018, pp. 1–5. research interests include network economics, game [39] A. Alsharoa, H. Ghazzai, A. Kadri, and A. E. Kamal, “Spatial and theory, wireless communications, edge computing, temporal management of cellular HetNets with multiple solar powered and distributed machine learning. drones,” IEEE Trans. Mobile Comput., vol. 19, no. 4, pp. 954–968, Apr. 2020. [40] T. Long, M. Ozger, O. Cetinkaya, and O. B. Akan, “Energy neutral Internet of drones,” IEEE Commun. Mag., vol. 56, no. 1, pp. 22–28, S. M. Ahsan Kazmi received the Ph.D. degree from Jan. 2018. Kyung Hee University in 2017. From 2017 to 2018, [41] S. Sekander, H. Tabassum, and E. Hossain, “On the performance he was also a Post-Doctoral fellow with the Depart- of renewable energy-powered UAV-assisted wireless communica- ment of Computer Science and Engineering, Kyung tions,” 2019, arXiv:1907.07158. [Online]. Available: https://arxiv.org/ Hee University, South Korea. From 2018 to 2020, abs/1907.07158 he was an Assistant Professor with the Institute [42] W.-S. Chan and L. Zhang, “Direct derivation of finite-time ruin of Information Security and Cyber-Physical System, probabilities in the discrete risk model with exponential or geomet- Innopolis University, Russia. He is currently a Senior ric claims,” North Amer. Actuarial J., vol. 10, no. 4, pp. 269–279, Lecturer in data science with the Department of Oct. 2006. Computer Science and Creative Technologies, Uni- [43] A. Manzoor, N. H. Tran, W. Saad, S. M. A. Kazmi, S. R. Pandey, and versity of the West of England, Bristol. His research C. S. Hong, “Ruin theory for dynamic spectrum allocation in LTE-U interests include the applying analytical techniques of optimization, machine networks,” IEEE Commun. Lett., vol. 23, no. 2, pp. 366–369, Feb. 2019. learning, and game theory to radio resource management for future networks. [44] M. Davis et al., Insurance Risk and Ruin. Cambridge, U.K.: Cambridge He received the Best KHU Thesis Award in engineering in 2017 and several Univ. Press, Oct. 2005. best paper awards from prestigious conferences. [45] W. Jaafar and H. Yanikomeroglu, “Dynamics of laser-charged UAVs: A battery perspective,” IEEE Internet Things J., early access, Dec. 29, 2020, doi: 10.1109/JIOT.2020.3048087. [46] A. Al-Hourani, S. Kandeepan, and S. Lardner, “Optimal LAP altitude Nguyen H. Tran (Senior Member, IEEE) received for maximum coverage,” IEEE Wireless Commun. Lett., vol. 3, no. 6, the B.S. degree in electrical and computer engi- pp. 569–572, Dec. 2014. neering from HCMC University of Technology [47] Evolved Universal Terrestrial Radio Access (E-UTRA); Radio Resource in 2005 and the Ph.D. degree in electrical and Control (RRC); Protocol Specification, document TS 36.331, Version computer engineering from Kyung Hee University 14.2.2, 3GPP, Apr. 2017. [Online]. Available: https://portal.3gpp.org/ in 2011. From 2012 to 2017, he was an Assistant desktopmodules/Specifications/SpecificationDet%ails.aspx?specification Professor with the Department of Computer Science Id=2440 and Engineering, Kyung Hee University. Since 2018, [48] U. Challita, W. Saad, and C. Bettstetter, “Deep reinforcement learn- he has been with the School of Computer Science, ing for interference-aware path planning of cellular-connected UAVs,” The University of Sydney, where he is currently in Proc. IEEE Int. Conf. Commun. (ICC), Kansas City, MO, USA, a Senior Lecturer. His research interests include May 2018, pp. 1–7. distributed computing, machine learning, and networking. He received the [49] M. N. Boukoberine, Z. Zhou, and M. Benbouzid, “A critical review Best KHU Thesis Award in engineering in 2011 and several best paper on unmanned aerial vehicles power supply and energy management: awards, including IEEE ICC 2016 and ACM MSWiM 2019. He receives the Solutions, strategies, and prospects,” Appl. Energy, vol. 255, Dec. 2019, Korea NRF Funding for Basic Science and Research from 2016 to 2023 and Art. no. 113823. ARC Discovery Project from 2020 to 2023. He was an Editor of IEEE [50] F. Qamar, K. Dimyati, M. N. Hindia, K. A. Noordin, and I. S. Amiri, TRANSACTIONS ON GREEN COMMUNICATIONS AND NETWORKING from “A stochastically geometrical Poisson point process approach for the 2016 to 2020, and the Associate Editor of IEEE JOURNAL OF SELECTED future 5G D2D enabled cooperative cellular network,” IEEE Access, AREAS IN COMMUNICATIONS from 2020 to 2021 in the area of distributed vol. 7, pp. 60465–60485, May 2019. machine learning/federated learning.

Authorized licensed use limited to: Kyunghee Univ. Downloaded on June 17,2021 at 05:23:30 UTC from IEEE Xplore. Restrictions apply. 3956 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 69, NO. 6, JUNE 2021

Walid Saad (Fellow, IEEE) received the Ph.D. Choong Seon Hong (Senior Member, IEEE) degree from the University of Oslo in 2010. He is received the B.S. and M.S. degrees in electronic currently a Professor with the Department of Electri- engineering from Kyung Hee University, Seoul, cal and Computer Engineering, Virginia Tech, where South Korea, in 1983 and 1985, respectively, and the he also leads the Network sciEnce, Wireless, and Ph.D. degree from Keio University, Minato, Japan, Security (NEWS) Laboratory. His research inter- in 1997. In 1988, he joined Korea Telecom Co., Ltd., ests include wireless networks, machine learning, where he was involved in broadband networks as a game theory, security, unmanned aerial vehicles, Member of the Technical Staff. In 1993, he joined cyber-physical systems, and network science. Keio University. He was with the Telecommuni- Dr. Saad was a recipient of the NSF CAREER cations Network Laboratory, Korea Telecom, as a Award in 2013, the AFOSR Summer Faculty Fel- Senior Member of Technical Staff and the Director lowship in 2014, and the Young Investigator Award from the Ofﬁce of Naval of the Networking Research Team until 1999. Since 1999, he has been a Research (ONR) in 2015. He has authored or coauthored ten conference best Professor with the Department of Computer Science and Engineering, Kyung paper awards at WiOpt in 2009, ICIMP in 2010, IEEE WCNC in 2012, Hee University. His research interests include future Internet, ad hoc networks, IEEE PIMRC in 2015, IEEE SmartGridComm in 2015, EuCNC in 2017, network management, and network security. IEEE GLOBECOM in 2018, IFIP NTMS in 2019, IEEE ICC in 2020, and Prof. Hong is a member of the ACM, IEICE, IPSJ, KIISE, KICS, KIPS, IEEE GLOBECOM in 2020. He was also a recipient of the 2015 Fred W. and OSIA. He has served as the general chair, a TPC chair/member, or an Ellersick Prize from the IEEE Communications Society, of the 2017 IEEE organizing committee member for international conferences, such as NOMS, ComSoc Best Young Professional in Academia Award, of the 2018 IEEE IM, AP-NOMS, E2EMON, CCNC, ADSN, ICPP, DIM, WISA, BcN, TINA, ComSoc Radio Communications Committee Early Achievement Award, and SAINT, and ICOIN. In addition, he is currently an Associate Editor of IEEE of the 2019 IEEE ComSoc Communication Theory Technical Committee. TRANSACTIONS ON NETWORK AND SERVICE MANAGEMENT, International He has coauthored the 2019 IEEE Communications Society Young Author Journal of Network Management,andJournal of Communications and Net- Best Paper. From 2015 to 2017, he was named the Stephen O. Lane Junior works, and an Associate Technical Editor of IEEE Communications Magazine. Faculty Fellow at Virginia Tech and, in 2017, he was named College of Engineering Faculty Fellow. He received the Dean’s award for Research Excellence from Virginia Tech in 2019. He currently serves as an Editor for IEEE TRANSACTIONS ON MOBILE COMPUTING and IEEE TRANSACTIONS ON COGNITIVE COMMUNICATIONS AND NETWORKING. He is an Editor- at-Large for IEEE TRANSACTIONS ON COMMUNICATIONS. He is an IEEE Distinguished Lecturer.

Authorized licensed use limited to: Kyunghee Univ. Downloaded on June 17,2021 at 05:23:30 UTC from IEEE Xplore. Restrictions apply.